Agent overview
The Agent is an AI assistant built into Tensor Cloud. Describe what you need in your own words, in Japanese or English, and it works through the data in your workspace to answer. Ask for a quick count, a table for a report, or a comparison of two simulation versions, whenever you need it.
We keep adding to what the agent can do, so this page does not list every skill. The quickest way to find out whether it can help with something is to ask. If an answer falls short, tell us. We use your feedback to decide what to build next.
The agent can make mistakes. Review its answers before you rely on them.
Starting a task
Click Agent at the top of the side navigation. Each conversation with the agent is called a task. Type what you need into the message box and press Enter, or click the send button. A message can be up to 10,000 characters long.
When the answer arrives, you can continue in the same task. Ask a follow-up question, narrow the scope, or ask for the same result in another form, such as a table. The agent takes the earlier messages in the task into account. It replies in the language you write in, and you can switch language at any time.
Keyboard shortcuts
- Shift+Enter: Starts a new line instead of sending.
- Escape: Stops the agent while it is working.
- Up arrow: In an empty message box, brings back messages you sent recently. Down arrow moves forward again.
While the agent works
A section above the answer shows what the agent is doing. Click it to expand the steps. To end the run early, click the stop button that replaces the send button, or press Escape. A stopped answer is marked Stopped.
You can type your next message while the agent works, but you can send it only once the current answer has finished. To change direction sooner, stop the agent first.
You do not have to wait for a task to finish. Start another task or work elsewhere in Tensor Cloud: running tasks continue in the background, and the Agent item in the side navigation shows how many of them are still running.
What to ask
The agent reads the data in the workspace you have open, such as your projects, simulations, contracts, SPVs, balancing groups, and the events recorded on your projects. It can also explain how Tensor Cloud works and answer questions about the Japanese power market.
The prompts below are starting points. Try your own, and word them the way you would ask a colleague.
| If you want to | Try something like |
|---|---|
| Find records | Which projects in Kyushu combine solar and battery? |
| Get totals | How many projects do we have in each grid area, and what is their total AC capacity? |
| Understand a simulation | Compare the two latest simulation versions of this project. What changed, and how did the economics move? |
| Follow up on operations | Which projects have open incidents, and which of those affect availability? |
| Review contracts | List our active PPAs with their offtakers. |
| Clean up data | Are any of our projects recorded more than once? |
| Check the weather | What was the solar irradiance at this project last week? |
| Learn about the market | How does imbalance settlement work for a FIP project? |
| Learn about Tensor Cloud | How do I share a simulation with an external stakeholder? |
A few habits lead to better answers:
- Say which projects, contracts, or period you mean. Phrases like "last quarter" or "since April" work.
- Ask for the format you want, such as a table, a short summary, or a list sorted by capacity.
- If the first answer misses the point, rephrase or add detail in the same task.
When an answer mentions a project or another record, it can link to that record in Tensor Cloud.
Context specific to your business
By default, the agent knows Tensor Cloud and the data stored in your workspace. It can also be given context specific to your organization: how your business processes run, how your integrations with other systems are set up, and details about your portfolio that do not fit into the platform's fields. With this context, the agent can answer in your own terms and help with questions that depend on how your company works.
Tensor sets up this context together with you, and it applies to every member of the workspace. Anything in it can appear in the agent's answers to any member, so it should only hold information the whole workspace is allowed to see. To add or update the context for your workspace, contact your account manager or use the Get support button in the top navigation bar.
Attaching files
You can attach files to a message and ask the agent about them, for example to summarize a report or pull figures out of a spreadsheet. Click the attach button in the message box, paste files into it, or drag them onto the conversation.
The agent accepts PDF, CSV, XLSX, DOCX, and MD files of up to 100 MB each, and up to 100 files per message. You can send once every upload has finished. After you send, a file shows Reading... while the agent reads it, and Couldn't read with a reason if that fails, for example when the file is password-protected.
The files stay with the task. Open the Files panel from the conversation header, or press the ] key, to preview or download them later.
File attachments are enabled per workspace, so you may not see them yet. Contact us with the Get support button if you need them.
Approving changes
The agent does not change data in your workspace without asking. You can ask it to create tags, for example. When a task involves a change, the agent shows a card for each action it proposes, with the values it intends to set, and waits for you. Click Approve or Deny. If it proposes several actions at once, decide on each one and then click Submit decisions.
The task is on hold until you decide. If you send a new message instead, the agent asks you to confirm and then cancels the proposed actions.
Managing tasks
Your tasks are listed beside the conversation, newest first. Only you can see them; other members of the workspace have their own list. Each workspace has its own tasks, so switching workspaces switches the list. Click New task to start a fresh conversation, or use the search box to find an earlier one by title.
To tidy up the list, hover over a task, open its ⋯ menu, and select Archive. Archived tasks are read-only. To continue one, open the archive and select Unarchive.
Giving feedback
Below each answer you can copy the text and rate it. Good response records your rating right away. Bad response opens a short form where you choose what went wrong (wrong, incomplete, too long, didn't follow instructions, bad action, or other) and describe what the agent should have done. Both fields are optional.
Please use these buttons freely. The ratings and comments go to the Tensor team and help us decide what to improve. If you have a bigger idea, such as a type of question you would like the agent to handle or a task you would like it to take over, tell your account manager or use the Get support button described on the help page.
Security and privacy
The agent works only inside the workspace you have open and can access only that workspace's data. To ask about another workspace, switch to it first. The agent is built for Tensor Cloud, your portfolio, and the energy market, and declines requests outside those topics.
Errors and connection problems
- Failed to send: A message that could not be sent is marked this way, with a Retry button. A notification explains why, for example Rate limit reached. Please wait and try again.
- The agent is still working on this task: You sent a message before the current answer finished. Your message goes back into the message box; send it again once the agent is done.
- The assistant is busy right now.: The agent is overloaded. Click Try again. If it is still busy after a few attempts, wait a moment and send your message again.
- Connection unstable: The live connection keeps dropping, so updates may be delayed. The banner clears once the connection recovers.
- Connection lost: The conversation could not be restored from the server, so what you see may be out of date. Check your network or reload the page.
If a problem persists, see Help.